You are on page 1of 11

Communications in Algebra

ISSN: 0092-7872 (Print) 1532-4125 (Online) Journal homepage: https://www.tandfonline.com/loi/lagb20

Involutions on tensor products of quaternion


algebras

Demba Barry

To cite this article: Demba Barry (2019) Involutions on tensor products of quaternion algebras,
Communications in Algebra, 47:8, 3229-3238, DOI: 10.1080/00927872.2018.1555834

To link to this article: https://doi.org/10.1080/00927872.2018.1555834

Published online: 11 Feb 2019.

Submit your article to this journal

Article views: 55

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=lagb20
COMMUNICATIONS IN ALGEBRAV
R

2019, VOL. 47, NO. 8, 3229–3238


https://doi.org/10.1080/00927872.2018.1555834

Involutions on tensor products of quaternion algebras


Demba Barrya,b
a
DER de Mathematiques et Informatique, Universite des Sciences des Techniques et des Technologies de
Bamako, Bamako, Mali; bDepartment Wiskunde–Informatica, Universiteit Antwerpen, Antwerpen, Belgium

ABSTRACT ARTICLE HISTORY


We study possible decompositions of totally decomposable algebras with Received 12 August 2016
involution, that is, tensor products of quaternion algebras with involution. Revised 10 September 2018
In particular, we are interested in decompositions in which one or several Communicated by Adrian
factors are the split quaternion algebra M2 ðFÞ, endowed with an orthog- Wadsworth
onal involution. We construct examples of algebras isomorphic to a tensor KEYWORDS
product of quaternion algebras with k split factors, endowed with an invo- Central simple algebra;
lution which is totally decomposable, but does not admit any decompos- involution; quaternion
ition with k factors M2 ðFÞ with involution. This extends an earlier result of algebra; valuation
Sivatski where the algebra considered is of degree 8 and index 4, and
endowed with some orthogonal involution. 2010 MATHEMATICS
SUBJECT CLASSIFICATION
Primary 16W10; Secondary
16K20; 16W60; 11E39

1. Introduction
Let F be a field of characteristic different from 2 and A a central simple F-algebra of degree 8
and exponent 2. Tignol showed that M2 ðAÞ is isomorphic to a tensor product of four quaternion
algebras (see for instance [6, Theorem 5.6.38]). Moreover, if A is indecomposable (i.e., A does
not contain any quaternion subalgebra, see [1]), one may check that M2 ðAÞ does not admit any
decomposition as a tensor product of quaternion algebras in which one of the quaternion factors
is split, that is isomorphic to M2 ðFÞ. In this paper, we are interested in the analogous question
for algebras with involution.
A central simple F-algebra A of exponent 2 is called totally decomposable if A is isomorphic to
a tensor product of quaternion algebras over F, that is, A ’ Q1 F    F Qm , for some quaternion
algebras Qi. The quaternion algebras Q1 ; :::; Qm are called factors of A. Moreover, we say that A
is totally decomposable with k split factors if it admits a decomposition in which k factors are
split quaternion algebras. This implies that the co-index of A, which is the ratio degA=indA is a
multiple of 2k ; but as explained before, this co-index condition is not sufficient for a totally
decomposable algebra to admit a decomposition with k split factors. Assume now that A is
endowed with an F-linear involution r. We say that ðA; rÞ is totally decomposable if ðA; rÞ ’
ðQ1 ; r1 ÞF    F ðQm ; rm Þ and totally decomposable with k split factors if it admits such a
decomposition with k factors isomorphic to M2 ðFÞ endowed with some F-linear involution.
Given a symmetric or alternating form u ¼ ðV; bÞ over F, following Becher’s notation [2], we
denote by Adu;F the adjoint split F-algebra with (orthogonal or symplectic) involution
ðEndF ðVÞ; adu Þ where adu is the involution on EndF ðVÞ that is adjoint to u. Hence, by definition
of Pfister forms, the algebra with involution ðA; rÞ is totally decomposable with k split factors if

CONTACT Demba Barry Barry.Demba@gmail.com DER de Mathematiques et Informatique, Universite des Sciences des
Techniques et des Technologies de Bamako, Bamako, Mali.
ß 2018 Taylor & Francis Group, LLC
3230 D. BARRY

and only if there exist quaternion F-algebras with involution ðQ01 ; r01 Þ; :::; ðQ0mk ; r0mk Þ such that
   
ðA; rÞ ’ Q01 ; r01 F    F Q0mk ; r0mk F Adp;F
where p is a (symmetric or alternating) k-fold Pfister form over F. The main problem we are
interested in is the following:

Question. Let ðA; rÞ be a totally decomposable F-algebra of degree 2m with involution such that
A is Brauer equivalent to a product of ðmkÞ quaternion algebras over F (i.e., A is of co-index 2k
and totally decomposable with k split factors). Does it follow that ðA; rÞ is it totally decomposable
with k split factors?
We give a complete answer to this question in the orthogonal case. Let ðA; rÞ be a totally
decomposable algebra of degree 2m with orthogonal involution. If A is split or of index 2 (that is,
co-index 2m1 ), Becher gives a positive answer to the problem in [2] (see also Proposition 2.2).
Notice that in the particular case where A is split, the question is known in the literature as the
Pfister Factor Conjecture (see [12, Chapter 9]). In higher index, the answer is negative in general
for orthogonal involution: in [13, Proposition 5], Sivatski constructs an algebra with orthogonal
involution of degree 8 and co-index 2 which is totally decomposable, but not totally decompos-
able with a split factor. Starting from Sivatski’s example, we construct examples of algebras with
orthogonal involution, of index 2r and co-index 2k , where r  2 and k  1 are arbitrary, that are
not totally decomposable with k split factors as algebras with involution, even though the under-
lying algebra does admit a total decomposition with k split factors (see Corollary 4.9). The way
how Sivatski’s example is extended is based on a construction (and related results) over iterated
power series fields.
A partial answer to the main question is given in the symplectic case. If A is split (in this case
the involution is hyperbolic), we explain in Proposition 2.2 why the question admits a positive
answer. In index 2, a positive answer is known by Becher [2]. In degree 8 and index 4, we prove
that ðA; rÞ is totally decomposable with 1 split factor (see Proposition 2.2). Finally, in index 2r
and co-index 2k for arbitrary r  3 and k  1, we give examples of ðA; rÞ such that A is totally
decomposable with k split factors but ðA; rÞ is not totally decomposable with k split factors (see
Corollary 4.9).
Throughout this paper the characteristic of the base field F is assumed to be different from 2
and all algebras are associative and finite-dimensional over their centers. We will use standard
terminology and notation from the theory of finite-dimensional algebras, the theory of valuations
on division algebras and the theory of involutions on central simple algebras. For these, as well as
background information, we refer the reader to [10] and [7].

2. Known results
In this section we collect known results concerning the main question. If ðA; rÞ is isotropic or
hyperbolic, then the answer to the question is always positive:
Lemma 2.1. Let ðA; rÞ be a totally decomposable algebra with involution over F. If r is isotropic,
then it is hyperbolic.
Moreover, if A is Brauer equivalent to a product of quaternion algebras ri¼1 Qi with 2r <degðAÞ
and ðA; rÞ is isotropic, then ðA; rÞ ’ ri¼1 ðQi ; ci Þ  Adp;F for a hyperbolic (symmetric or alternat-
ing) Pfister form p, where ci is the canonical involution on Qi.
Proof. A totally decomposable isotropic involution is hyperbolic by [4, Corollary 3.2].
For the second part, one takes the canonical involution ci on Qi for each i and a hyperbolic k-
Pfister form p of the relevant type (symmetric or alternating) on the split factors. w

We have a positive answer in the following cases.


COMMUNICATIONS IN ALGEBRAV
R
3231

Proposition 2.2. Let A be a central simple algebra over F and assume that ðA; rÞ is totally
decomposable.

(i) If A is split and of degre 2m , then ðA; rÞ ’ Adp;F where p is a (symmetric or alternating) m-
fold Pfister form over F, that is, ðA; rÞ is totally decomposable with m split factors.
(ii) If A is of degree 2m and co-index 2m1 (i.e., A is Brauer equivalent to a quaternion algebra
H), then ðA; rÞ ’ ðH; rH ÞF Adp0 ;F for some involution rH on H and some ðm1Þ-fold
Pfister form p0 over F.
(iii) If A is of degre 8, and co-index 2 and r symplectic, then ðA; rÞ is totally decomposable with
1 split factor.

Proof. If r is isotropic or hyperbolic, then the result follows by Lemma 2.1. We assume that r is
anisotropic. The statements in (i) and (ii) are due to Becher [2].
(iii) The index of A is 4. Since ðA; rÞ is totally decomposable, the cohomological invariant
DðA; rÞ, defined in [5], is zero. Therefore it readily follows by [3, Theorem 8] that ðA; rÞ is totally
decomposable with 1 split factor. w

For orthogonal involutions in degree 8 and co-index 2, the main question has a negative
answer: let A be a central simple F-algebra of degree 8 and co-index 2 and let r be an anisotropic
involution on A. If r is orthogonal, it follows from [9, Theorem. 1.1] that ðA; rÞ ’ ðD; hÞF Adu;F
where ðD; hÞ is some biquaternion F-algebra with orthogonal involution and u is a binary quad-
ratic form over F. Sivatski proved in [13] that the involution ðD; hÞ need not to be decomposable
in general. More precisely, he produced an example where ðA; rÞ does not admit any decompos-
ition into a tensor product of three quaternion algebras with involution with a split factor. Since
this is the starting point of our examples, we recall his construction.
Proposition 2.3 ([13, Proposition 5]). Let D be a biquaternion division algebra over a field k of
charðkÞ 6¼ 2 and let h be an orthogonal involution on D with nontrivial discriminant
s ¼ discðhÞ 6¼ 1. Set F ¼ kðxÞ, where x is an indeterminate over k and consider the algebra with
orthogonal involution
ðA0 ; r0 Þ ¼ ðD; hÞk Adhhx2 sii;F :

(1) A0 is of degree 8 and co-index 2. Moreover, A0 is totally decomposable with 1 split factor.
(2) ðA0 ; r0 Þ is totally decomposable.
(3) ðA0 ; r0 Þ is not totally decomposable with 1 split factor.
(4) The involution r0 is anisotropic.

Proof. (1) The statement follows from the definition of A0 .


For (2) and (3), see [13, Proposition 5].
(4) Assume that r0 is isotropic. Since ðA0 ; r0 Þ is totally decomposable and D is a biquaternion
algebra, it follows by Lemma 2.1 that ðA0 ; r0 Þ ’ ðQ1 ; c1 Þ  ðQ2 ; c2 Þ  Adh1;1i;F , for some quatern-
ion k-algebras Q2, Q3. That is impossible by part (3). Therefore r0 is anisotropic. w

3. Main result
The main result of this paper is the following:
Theorem 3.1. Let ðS; qÞ and ðS0 ; hÞ be F-algebras with anisotropic involutions. Let a 2 F  ; K ¼
FððtÞÞ and ðH; Þ a K-quaternion algebra with anisotropic orthogonal involution such that
ðS; qÞF Adhhatii;K ’ ðS0 ; hÞF ðH;  Þ:
3232 D. BARRY

Then

(i) ðH; Þ ’ Adhha0 tii;K for some a0 2 F  ,


(ii) ðS; qÞ ’ ðS0 ; hÞ.

For the proof of this theorem, we need the following result.


Lemma 3.2. Let ðS; qÞ be an F-algebra with involution, a 2 F and K ¼ FððtÞÞ. Then ðS; qÞ is
anisotropic if and only if ðS; qÞF Adhhatii;K is anisotropic.

Proof. Let D be the underlying division F-algebra of S, that is, S is an endomorphism algebra of
some n-dimensional D-vector space. The involution q on S is adjoint to a nondegenerate hermit-
ian form h with respect to some involution J on D. Hence, we may write
ðS; qÞF Adhhatii;K ’ ðEndDK V; adhhhatii Þ, where DK is endowed with the involution J  id and V
is a 2n-dimensional DK-vector space. Note that adhhhatii is anisotropic if and only if h  hhatii is
anisotropic.
Assume that q is anisotropic, that is, h is anisotropic. Let h ¼ hd1 ; :::; dn i, with d1 2 D and
Jðdi Þ ¼ di , a diagonalization of h. One has
h  hhatii ¼ hd1 ; :::; dn ; ad1 t; :::; adn ti:
Suppose h  hhatii is isotropic. There exist x1 ; :::; xn ; y1 ; :::; yn 2 DK such that
 
J ðx1 Þd1 x1 þ    þ J ðxn Þdn xn ¼ at J ðy1 Þd1 y1 þ    þ J ðyn Þdn yn : (3.1)
Notice that the t-adic valuation of K extends uniquely to a valuation v on D. We first show
that
   
v J ðx1 Þd1 x1 þ    þ J ðxn Þdn xn ¼ 2min vðx1 Þ; :::; vðxn Þ :
Set ‘ ¼ minðvðx1 Þ; :::; vðxn ÞÞ and assume that vðJðx1 Þd1 x1 þ    þ Jðxn Þdn xn Þ>2‘. We have
vðJðx1 t‘ Þd1 x1 t‘ þ    þ Jðxn t ‘ Þdn xn t‘ Þ>0 with vðxi t ‘ Þ  0 for 1  i  n and vðxj t‘ Þ ¼ 0 for
some j. Therefore the residue
   
J x1 t‘ d1 x1 t‘ þ    þ J xn t‘ dn xn t ‘ ¼ 0
is zero. It then follows that the form h ¼ hd1 ; :::; dn i is isotropic. This contradicts the hypothesis
on h. Thus we conclude that vðJðx1 Þd1 x1 þ    þ Jðxn Þdn xn Þ ¼ 2‘. Likewise,
 
v J ðy1 Þd1 y1 þ    þ J ðyn Þdn yn ¼ 2minðvðy1 Þ; :::; vðyn ÞÞ
is even. Hence, the equality (3.1) implies that 2‘ is odd since
    
v at J ðy1 Þd1 y1 þ    þ J ðyn Þdn yn ¼ 1 þ v J ðy1 Þd1 y1 þ    þ J ðyn Þdn yn :
This is impossible. Therefore h  hhatii is anisotropic.
Conversely, since h is a subform of h  hhatii, if h is isotropic then h  hhatii is isotropic. The
proof is complete. w

Let A be a central simple FððtÞÞ-algebra and A0 the division algebra Brauer equivalent to A.
 0 : F ¼
The algebra A is called unramified with respect to the t-adic valuation on FððtÞÞ if ½A
 
½A0 : FððtÞÞ and the center of A 0 is F, where A 0 is the residue of A0.
Proof of Theorem 3.1. For simplicity, we set
ðT; gÞ ¼ ðS; qÞF Adhhatii;K and ðT 0 ; g0 Þ ¼ ðS0 ; hÞF ðH;  Þ:
(i) Since S and S0 are unramified with respect to the t-adic valuation on K ¼ FððtÞÞ and since
Adhhatii;K is split, it follows that H is unramified, hence defined over F. We first show that ðT; gÞ
COMMUNICATIONS IN ALGEBRAV
R
3233

is not defined over F. As in the proof of Lemma 3.2, we write ðT; gÞ ’ ðEndDK V; adhhhatii Þ,
where DK is endowed with the involution J  id. By Lemma 3.2, the form h  hhatii is aniso-
tropic because h is anisotropic. Since no scalar multiple of h  hhatii has a trivial residue form,
ðT; gÞ is not defined over F. It also follows that ðH; Þ is not defined over F. As it is shown in
Corollary 4.2, the only possibility for having that H is defined over F but ðH; Þ is not is that
ðH; Þ ’ Adhha0 tii;K for some a0 2 F  .
(ii) Since ðT; gÞ ’ ðT 0 ; g0 Þ by hypothesis, it follows by part (i) that ðT 0 ; g0 Þ ’ ðS0 ; hÞF Adhha0 tii;K .
Let h0 be a Hermitian form with respect to (D, J) corresponding to h on S0 . The isomorphism
ðT; gÞ ’ ðT 0 ; g0 Þ implies that there exists l 2 K  such that h  hhatii ’ hliðh0  hha0 tiiÞ. It follows
by [11, Satz 3.5] that there exist l0 ; l1 2 F such that h ’ h0  hl0 i and h  hai ’ h0  hl1 i. We
deduce that ðS; qÞ ’ ðS0 ; hÞ. That concludes the proof. w

4. Decomposition with arbitrarily large degree


The goal of this section is to give examples that extend Sivatski’s result to algebras with involu-
tion of arbitrarily large co-index and to algebras with involution of arbitrarily large index. Our
constructions involve Laurent power series fields. First, let us fix some notations: let F be a field,
for c; d 2 F  , we denote by ðc; d=FÞ the quaternion F-algebra generated by two elements i and j
satisfying i2 ¼ c; j2 ¼ d, and ij ¼ ji.
We start out by proving some preliminary results on quaternion algebras over K ¼ FððtÞÞ,
where t is an indeterminate over F.
Lemma 4.1. Let ðH; Þ be a quaternion algebra with involution over K ¼ FððtÞÞ. There exist i; j 2
H  such that j2 2 F  ; ij þ ji ¼ 0; ðiÞ ¼ 6i and ðjÞ ¼ 6j.
Proof. We first consider H without the involution . If H is split, one has H ’ ða; 1=FÞ  K ’
M2 ðKÞ ’ ðat; 1=KÞ for a 2 F  . So, we may suppose H is a division algebra. If H is defined over
F, then H ’ ða; b=FÞ  K for some a; b 2 F . Now, assume that H is not defined over F. For a
pure quaternion q in H, we have q2 2 K  . Up to scaling, we may assume that either q2 ¼ at for
some a 2 F or q2 ¼ b for some b 2 F  because
   
K  =K 2 ¼ a  K 2 ja 2 F [ at  K 2 ja 2 F
(see [8, Corollary VI.1.3]). Since we assume that H is not defined over F, the algebra H is either
isomorphic to H ’ ðat; b=KÞ or to H ’ ðat; bt=KÞ ’ ðat; ab=KÞ.
We consider the involution  on H. If  is symplectic, we have  ¼ c where c is the canonical
involution on H. Assume that  is orthogonal. By [7, (2.7)], the involution  has the form
IntðsÞ c for some invertible s 2 H with cðsÞ ¼ s. It follows by [7, (7.3)] that
discðÞ ¼ NrdðsÞ  K 2 2 K  =K 2 , where Nrd denotes the reduced norm. Therefore, either
discðÞ ¼ at  K 2 for some a 2 F  or discðÞ ¼ b  K 2 for some b 2 F  . If discðÞ ¼ at  K 2 ,
we choose i ¼ s. One has  ¼ IntðiÞ c. Let j 2 H  be such that j2 ¼ y 2 K  =K 2 and ij ¼ ji
(recall that such a j always exists). Up to scaling we may assume y 2 F or y 2 F   t. We have
the isomorphism ðH; Þ ’ ððat; b=KÞ; IntðiÞ cÞ if y ¼ b 2 F and ðH; Þ ’ ððat; bt=KÞ; IntðiÞ cÞ
if y has the form y ¼ bt. In the latter case, we may substitute ij1 for j and reduce to the case
where j2 2 F .
Likewise, if discðÞ ¼ b  K 2 for some b 2 F , we choose j ¼ s. So  ¼ IntðjÞ c. Let i 2 H 
be such that i2 2 K  =K 2 and ij ¼ ji. Since ðH; Þ is not defined over F, up to scaling we have
necessarily i2 ¼ at for some a 2 F  . This yields that ðH; Þ ’ ððat; b=KÞ; IntðjÞ cÞ. Notice that,
in all cases the generators i and j satisfy the required condition. w

Corollary 4.2. Let ðH; Þ be a quaternion algebra with involution over K ¼ FððtÞÞ. Assume that H
is defined over F and ðH; Þ is not defined over F. Then ðH; Þ ’ Adhhatii;K for some a 2 F .
3234 D. BARRY

Proof. Since H is defined over F, the algebra with involution ðH; Þ is defined over F when  is
the canonical involution or orthogonal of discðÞ 2 F   K 2 =K 2 . Hence, the hypothesis implies
that the involution  is orthogonal of discriminant discðÞ ¼ at  K 2 for some a 2 F  . Lemma
4.1 shows that there exist i; j 2 H  such that j2 ¼ b 2 F  ; ij þ ji ¼ 0; ðiÞ ¼ 6i and ðjÞ ¼ 6j.
Since ðH; Þ is not defined over F, up to scaling, we may suppose i2 ¼ at. So
ðH; Þ ’ ððat; b=KÞ; Þ. The algebra H being defined over F, we have necessarily j2 ¼ b 2 F 2 .
Therefore ðH; Þ ’ Adhhatii;K since discðÞ ¼ at  K 2 . w

Lemma 4.3. Let B be a biquaternion algebra over K ¼ FððtÞÞ and  a decomposable involution on B.
(a) Then ðB; Þ has a decomposition with one factor defined over F.
(b) If B has zero divisors, then ðB; Þ has a decomposition with one split factor and one (possibly
the same) factor defined over F.

Proof. We write ðB; Þ ’ ðH1 ; 1 ÞK ðH2 ; 2 Þ for some quaternion algebras with involution
ðH1 ; 1 Þ and ðH2 ; 2 Þ over K.

(a) If ðH1 ; 1 Þ or ðH2 ; 2 Þ is already defined over F, there is nothing to show. We may therefore
assume that ðH1 ; 1 Þ and ðH2 ; 2 Þ are not defined over F. It follows from Lemma 4.1, there
exist quaternion generators i1, j1 and i2, j2 of H1 and H2 respectively such that
i2‘ ¼ a‘ t; j2‘ ¼ b‘ , and ‘ ði‘ Þ ¼ 6i‘ and ‘ ðj‘ Þ ¼ 6j‘ for ‘ ¼ 1; 2. Let H10 ; H20
H1 K H2 be
quaternion subalgebras generated by t1 i1  i2 ; j1  1 and 1  i2 ; j1  j2 respectively. In fact,
H10 ’ ða1 a2 ; b1 =FÞF K and H20 ’ ða2 t; b1 b2 =KÞ. We easily check that H10 and H20 are stable
under 1  2 . Denote by  0‘ (for ‘ ¼ 1; 2) the restriction of 1  2 to H 0‘ . We have the iso-
morphism
   
ðH1 ; 1 ÞK ðH2 ; 2 Þ ’ H 01 ;  01 K H 02 ;  02
with ðH 01 ;  01 Þ defined over F.
(b) Since B has a zero divisor, it follows from Proposition 2.2 (i) that ðB; Þ has a decompos-
ition with one split factor. Analogous computations on quaternion generators, as in part
(a), show that ðB; Þ has a decomposition with one split factor and one (possibly the same)
factor defined over F. This concludes the proof. w

Corollary 4.4. Let ðA; rÞ be an algebra with involution of degree 2n over K ¼ FððtÞÞ. Let s 2 N
with 0  s<n. Assume that ðA; rÞ is not defined over F and totally decomposable with ðs þ 1Þ split
factors. Then ðA; rÞ ’ ðS0 ; r0 ÞF ðH; Þ for a quaternion algebra with involution ðH; Þ not defined
over F and a totally decomposable algebra with involution ðS0 ; r0 Þ with s split factors. Moreover,
there exists a choice of ðS0 ; r0 Þ in which each factor is defined over F.

Proof. Starting with a decomposition with s þ 1 split factors, one first rearranges the non-split fac-
tors such that at most one of them is not defined over F. Then we use repeatedly (s þ 1 times)
part (b) of Lemma 4.3 to improve at each step a decomposition of a biquaternion algebra with
involution involving a split factor into a decomposition with a split factor defined over F. w

Lemma 4.5. Let H and H 0 be quaternion algebras over K ¼ FððtÞÞ such that HK H 0 is defined
over F. Then either H and H 0 are defined over F, or there exist a; b; b0 2 F  such that H ’
ða; bt=KÞ and H 0 ’ ða; b0 t=KÞ.

 0
Proof. If H is not defined over F, then we may write H ’ ða; bt=KÞ with
pffiffiffi a; b 2 F . Since HK H
0
is defined over F, it then follows that H becomes unramified over Fð aÞ. This is only possible if
H 0 ’ ða; b0 t=KÞ for some b0 2 F  . w
COMMUNICATIONS IN ALGEBRAV
R
3235

Consider the fields


F1 ¼ Fððt1 ÞÞ and F2 ¼ Fððt1 ÞÞððt2 ÞÞ;
where t1, t2 are independent indeterminates over F, and the quaternion algebra
Q ¼ ðat1 ; bt2 =F2 Þ
over F2 for a; b 2 F  . Let S be a central simple algebra over F and let q be an involution on S.
Let q0 be any involution of orthogonal or symplectic type on Q and let
ðS1 ; rÞ ¼ ðS; qÞF Adhhat1 ii;F1 ; ðS2 ; sÞ ¼ ðS; qÞF ðQ; q0 Þ:

Proposition 4.6. Assume that ðS; qÞ is totally decomposable of degree 2n and let s be a positive
integer with s  n. Then:

(1) The algebra with involution ðS1 ; rÞ is a totally decomposable F1-algebra with involution of
degree 2nþ1 and of same type and index as ðS; qÞ. Moreover, ðS1 ; rÞ is totally decomposable
with ðs þ 1Þ split factors if and only if ðS; qÞ is totally decomposable with s split factors.
(2) The algebra with involution ðS2 ; sÞ is a totally decomposable F2-algebra with involution of
degree 2nþ1 and with index indS2 ¼ 2  indS. Moreover, ðS2 ; sÞ is totally decomposable with s
split factors if and only if ðS; qÞ is totally decomposable with s split factors.

Proof. (1) Notice that indS ¼ ind SF1 and S1 ¼ SF M2 ðF1 Þ. We have ind S1 ¼ ind S; degS1 ¼
2nþ1 and the involutions q and r ¼ q  adhhat1 ii are of same type since adhhat1 ii is orthogonal. If
ðS; qÞ is totally decomposable with s split factors, then ðS1 ; rÞ is obviously decomposable with
ðs þ 1Þ split factors. Conversely, suppose ðS1 ; rÞ is totally decomposable with ðs þ 1Þ split factors.
If r is isotropic then q is isotropic by Lemma 3.2. So ðS; qÞ is hyperbolic since it is totally decom-
posable, Lemma 2.1. It then follows that ðS; qÞ is totally decomposable with s split factors because
the co-index of S is  s. Hence, we may assume that r is anisotropic. By Corollary 4.4, the alge-
bra with involution ðS1 ; rÞ admits a total decomposition, of the form ðS1 ; rÞ ’ ðS0 ; r0 ÞF ðH; ÞF1 ,
where ðS0 ; r0 Þ is a totally decomposable F-algebra with involution with s split factors and ðH; Þ is
a quaternion F1-algebra with involution. It follows from Theorem 3.1 that ðH; Þ ’ Adhha0 t1 ii;F1 for
some a0 2 F  and ðS0 ; r0 Þ ’ ðS; qÞ. Hence, ðS; qÞ is totally decomposable with s split factors since
ðS0 ; r0 Þ is totally decomposable with s split factors.
(2) Clearly, ðS2 ; sÞ ¼ ðS; qÞF ðQ; q0 Þ is of degree 2nþ1 and indS2 ¼ 2indS. By hypothesis the F-
algebra with involution ðS; qÞ is totally decomposable. Obviously, If ðS; qÞ is totally decomposable
with s split factors, then ðS2 ; sÞ is totally decomposable with s split factors. Conversely, suppose
the F2-algebra with involution ðS2 ; sÞ ¼ i¼1 nþ1
ðQi ; si Þ, where we set ðQ; q0 Þ ¼ ðQnþ1 ; snþ1 Þ, is
totally decomposable with s split factors. It follows by Corollary 4.4 that
nþ1 nþ1  

i¼1
ðQi ; si Þ ’ 
i¼1
Q0i ; s0i (4.1)

for F2-quaternion algebras with involution ðQ01 ; s01 Þ; :::; ðQ0nþ1 ; s0nþ1 Þ such that
8   0 0
> 0 0
< Q1 ; s1 ; :::; Qs ; ss are split;
Q01 ; s01 ; :::; Q0n1 ; s0n1 are defined over F;
: Q0 ; s0  is defined over F :
>
n n 1

In fact, we first use Corollary 4.4 for the field extension F2 =F1 and in the second step we use
the same for the field extension F1 =F.
Let Falg be an algebraic closure of F and F~2 ¼ Falg ððt1 ÞÞððt2 ÞÞ. Note that Qi and Q0i split over
F~2 for 1  i  n while ðQnþ1 ÞF~2 is the unique F~2 -quaternion division algebra. It follows that
3236 D. BARRY

Q0nþ1 ¼ ða0 t1 ; b0 t2 =F2 Þ for some a0 ; b0 2 F  . By applying similar manipulations as in the proof of
Lemma 4.3 to ðQ0n ; s0n Þ and ðQ0nþ1 ; s0nþ1 Þ, we may assume that ðQ0n ; s0n Þ is defined over F.
We claim that Qnþ1 ’ Q0nþ1 . Indeed, notice that the isomorphism (4.1) implies that
   
Qnþ1 F2 Q0nþ1 ni¼1 Qi F2 ni¼1 Q0i ;
where denotes Brauer equivalence. So Qnþ1 F2 Q0nþ1 is defined over F since the right side of this
Brauer equivalence is defined over F. Applying Lemma 4.5 to Qnþ1 F2 Q0nþ1 over the field exten-
sion F2 =F1 , it follows that Q0nþ1 ’ ðat1 ; ct2 =F2 Þ for some c of the form c 2 F  or c 2 t1 F . If c ¼
dt1 with d 2 F  , we obtain ðat1 ; ct2 =F2 Þ ¼ ðat1 ; dt1 t2 =F2 Þ ’ ðat1 ; adt2 =F2 Þ. So, we may always
suppose Q0nþ1 ¼ ðat1 ; ct2 =F2 Þ with c 2 F  . Hence
Qnþ1 F2 Q0nþ1 ðat1 ; bc=F2 Þ:
The algebra Qnþ1 F2 Q0nþ1 being defined over F, we have necessarily bc 2 F2 , which also
means that ðat1 ; bc=F2 Þ is split. Therefore, Qnþ1 F2 Q0nþ1 is a split biquaternion algebra, that is,
Qnþ1 ’ Q0nþ1 and the claim is proved.
We deduce that
ðS; qÞF ððQnþ1 ; snþ1 ÞF2 ðQnþ1 ; snþ1 ÞÞ ¼ ðS2 ; sÞF2 ðQnþ1 ; snþ1 Þ
!
nþ1 
’  Qi ; si F2 ðQnþ1 ; snþ1 Þ
0
i¼1

is totally decomposable with ðs þ 2Þ split factors. But the involution snþ1  snþ1 on
Qnþ1 F2 Qnþ1 is adjoint to the involution trace form of snþ1 (see [7, (4.8) or (11.1)]).
Computations show that this trace form is similar to hh6at1 ; 6bt2 ii, where the sign ± depends
on the type of snþ1 . Hence,
ðS; qÞF ððQnþ1 ; snþ1 ÞF2 ðQnþ1 ; snþ1 ÞÞ ’ ðS; qÞF Adhh6at1 ;6bt2 ii;F2
is totally decomposable with ðs þ 2Þ split factors. Applying twice part (1) yields that ðS; qÞ is
totally decomposable with s split factors. w

Notice that if ðS; qÞ is totally decomposable then the algebras with involution ðS1 ; rÞ and
ðS2 ; sÞ are totally decomposable. But we have:
Remark 4.7. Decompositions of the algebra ðS1 ; rÞ of Proposition 4.6 do not necessarily arise
from a decomposition of ðS; qÞ by multiplication with the factor Adhhat1 ii;F1 . For instance, the
arguments in the proof of Lemma 4.3 show that if ðS; qÞ is a quaternion algebra ðc; d=FÞ with its
canonical involution, then
ðS; qÞF Adhhat1 ii;F1 ’ ððcat1 ; d=F1 Þ;  ÞF1 ððat1 ; d=F1 Þ; cÞ
where c is the canonical involution and  an orthogonal involution with discriminant cat.
Likewise, decompositions of the algebra ðS2 ; sÞ of Proposition 4.6 do not necessarily arise from a
decomposition of ðS; qÞ.
Iterating the Laurent series construction, we apply Proposition 4.6 inductively to give negative
answers to the main question for algebras of arbitrarily large degree: let m  4 be an arbitrary
integer. Consider the field of iterated Laurent series in m – 3 indeterminates
F^ ¼ Fððt1 ÞÞ; :::; ððtm3 ÞÞ

Corollary 4.8. Let ðA0 ; r0 Þ be the algebra with orthogonal involution of Proposition 2.3 and let
 
^ r
A; ^ ¼ ðA0 ; r0 ÞF Adhht1 ;:::;tm3 ii;F^ :
COMMUNICATIONS IN ALGEBRAV
R
3237

Then:

(a) ^ is of degree 2m and co-index 2m2 .


A
(b) ^ is totally decomposable with ðm2Þ split factors.
A
(c) ^ r
ðA; ^ Þ is not totally decomposable with ðm2Þ split factors.

Proof. (a) and (b) follow immediately from the definition of A. ^


0 0
(c) For m ¼ 4, assume that ðA ; r ÞF Adhht1 ii;F1 is totally decomposable with 2 split factors. It
follows by Proposition 4.6 (1) that ðA0 ; r0 Þ is totally decomposable with 1 split factor; that is
impossible. The corollary follows by induction on m. w

Corollary 4.9. There exist totally decomposable algebras with involution of index 2r , where r  2
for orthogonal involutions and r  3 for symplectic involutions, and co-index 2k with k  1 that
are not totally decomposable with k split factors.

Proof. Let recall that Q ¼ ðat1 ; bt2 =F2 Þ for some a; b 2 F and q0 is an involution (of orthogonal
or symplectic type) on Q. Let ðA0 ; r0 Þ be the algebra with orthogonal involution of Proposition
2.3. The algebra A0 F Q is of degree 16 and co-index 2. Moreover, A0 F Q is totally decomposable
with 1 split factor by its definition. But ðA0 ; r0 ÞF ðQ; q0 Þ is not totally decomposable with 1 split
factor. Indeed, Assume that ðA0 ; r0 ÞF ðQ; q0 Þ is totally decomposable with 1 split factor. It follows
from Proposition 4.6 (2) that ðA0 ; r0 Þ is totally decomposable with 1 split factor; that is impos-
sible. Therefore ðA0 ; r0 ÞF ðQ; q0 Þ is not totally decomposable with 1 split factor.
One iterates this construction to get a totally decomposable algebra with involution of index 2r
and co-index 2 that is not totally decomposable with 1 split factor. Next, we apply Corollary 4.8
to obtain an arbitrary large degree while the index is fixed. This process provides algebras with
involution with the desired properties. w

Acknowledgments
I am very grateful to Jean-Pierre Tignol for triggering the idea presented in this paper and for his availability. I
would like to thank Anne Queguiner-Mathieu for inspiring discussion. I thank the anonymous referees for the
careful reading and for suggestions that improved the content and clarity for this article.

Funding
This work was supported by the French Agence Nationale de la Recherche (ANR) under reference ANR-12-BL01-
0005, the French Embassy in Bamako (grant no. 830592G (2014)), and the FWO Odysseus Programme (project
Explicit Methods in Quadratic Form Theory).

References
[1] Amitsur, S. A., Rowen, L. H., Tignol, J.-P. (1979). Division algebras of degree 4 and 8 with involution,
Isreal. Israel J. Math. 33:133–148.
[2] Becher, K. J. (2008). A proof of the Pfister factor conjecture. Invent. Math. 173(1): 1–6.
[3] Berhuy, G., Monsurr o, M., Tignol, J.-P. (2003). The discriminant of a symplectic involution. Pacific J.
Math. 209(2): 201–218.
[4] Black, J., Queguiner-Mathieu, A. (2014). Involutions, odd degree extensions and generic splitting. Enseign.
Math. 60(3): 377–395.
[5] Garibaldi, S., Parimala, R., Tignol, J.-P. (2009). Discriminant of symplectic involutions. Pure App. Math.
Quart. 5(1): 349–374.
[6] Jacobson, N. (1996). Finite-Dimensional Division Algebras over Fields. Berlin: Springer-Verlag.
3238 D. BARRY

[7] Knus, M.-A., Merkurjev, S. A., Rost, M., Tignol, J.-P. (1998). The Book of Involutions, Colloquium Publ.,
Vol. 44. Providence, RI: AMS.
[8] Lam, T. Y. (2005). Introduction to Quadratic Forms over Fields, Graduate Studies in Mathematics, Vol. 67.
Providence, RI: AMS.
[9] Masquelein, A., Queguiner-Mathieu, A., Tignol, J.-P. (2009). Quadratic forms of dimension 8 with trivial
discriminant and Clifford algebra of index 4. Arch. Math. 93(2): 129–138.
[10] Pierce, R. S. (1982). Associative Algebras, Graduate Texts in Mathematics, Vol. 88. New York: Springer-
Verlag.
[11] €ber lokalen k€
Scharlau, W. (1970). Klassifikation hermitescher formen u orpen. Math. Ann. 186(3): 201–208.
[12] Shapiro, D. B. (2000). Compositions of Quadratic Forms, Expositions in Mathematics, Vol. 33, Berlin: De
Gruyter.
[13] Sivatski, A. S. (2005). Applications of Clifford algebras to involutions and quadratic forms. Commun.
Algebra. 33(3): 937–951.

You might also like